Anthropic Blocks Fable 5 and Mythos 5: US Export Ban

Anthropic sharply criticizes the measure. The company emphasizes that, prior to the launch of Fable 5, it had implemented extensive safety measures and pursued a so-called defense-in-depth strategy. The protective mechanisms had been tested over thousands of hours together with the US government, the British AI safety initiative AISI, and private third-party organizations. In the company's view, a possible, narrowly limited jailbreak does not justify the withdrawal of a commercial model already used by hundreds of millions of people.

The dispute between Anthropic and Washington has a longer history. In February 2026, the Pentagon had classified the company as a "supply-chain risk," a category normally reserved for foreign adversaries or compromised technology providers. According to widespread reports, it was the first time an American company had been classified in this way, and the first time the designation had been made in response to a company's refusal to accept certain contractual terms. The contract negotiations between Anthropic and the Pentagon had previously failed over two red lines, among other things: Anthropic wanted to ensure that its AI would not be used for mass surveillance of US citizens or for autonomous weapons systems; the Pentagon insisted on use for all lawful purposes.

Background: Pentagon Classifies Anthropic as a Supply-Chain Risk

According to information from the news agency Axios, the trigger for the current block was a report in which another company claimed to have found a jailbreak for the underlying model line Mythos. The Department of Commerce then decided on the export control measure. Anthropic disputes the severity of the incident: its own experts had concluded that the report merely described a limited capability with which the AI could review certain program code and correct errors. Comparable functions are offered by other publicly available models, such as OpenAI's GPT-5.5, which are not subject to comparable export control restrictions.

Trigger: Report of an Alleged Jailbreak

The tech investor David Sacks, the former "AI czar" of the Trump administration and co-chair of the President's Council of Advisors on Science and Technology, provides insight into the sequence of events. Sacks continues: "Die Administration forderte [Anthropic-CEO Dario Amodei] auf, den Jailbreak entweder zu flicken oder das Modell vorerst vom Netz zu nehmen." As far as is known, the government had received information only about a "potential" jailbreak, not a verified, critical one. Moreover, the supposedly uncovered vulnerabilities had been "minimal." Sacks confirmed the role of a partner who was not named: "Ein hochgradig glaubwürdiger, vertrauenswürdiger Partner sowohl von Anthropic als auch der US-Regierung trat mit einem funktionierenden Jailbreak an uns heran," Sacks said.

Meanwhile, according to tech investor Sacks, the discussion has stalled. "Dario weigerte sich," as TechCrunch reports. Sacks stressed, however, that this did not have to mean permanent censorship: "Die Hoffnung der Administration ist nun, dass Anthropic das Sicherheitsproblem behebt, die Exportkontrolle aufgehoben wird und Fable wieder in den allgemeinen Release gehen kann." Because of this refusal, the government ultimately and "reluctantly" pulled the emergency brake and imposed a strict export ban. Anthropic has refused to disclose the details of these discussions, according to the company: "Wenn dies geschieht, teilen wir die Details dieser Diskussionen nicht."

Role of the Parties Involved: Sacks, Amazon, and Anthropic

Amazon, itself one of Anthropic's largest investors and closely intertwined with the AI company through its cloud division AWS, was tight-lipped when asked by the media: "Als führender Cloud-Anbieter, der eine große Anzahl von Kunden im privaten und öffentlichen Sektor bedient, ist es nicht ungewöhnlich, dass Regierungen unseren Rat zu potenziellen Sicherheitsrisiken suchen," an Amazon spokesperson said.

At Anthropic, there is disbelief over the government's heavy-handed approach and the support from Seattle. The company had staged Mythos for weeks as "too dangerous for the public" in order to generate exclusivity. Now the regulators in Washington took the startup at its word and responded to the intense "disaster marketing." The AI company emphasizes that it had implemented extensive safety measures before the launch of Fable 5, and according to its own statements is working to restore access to Fable 5 and Mythos 5 as quickly as possible.

The model "Fable 5," released only this week, is based on Mythos technology, but with disabled cybersecurity and biotechnology capabilities. Mythos 5 itself is a non-public full version intended exclusively for government agencies and selected corporate partners for hardening their systems. Mythos's ability to detect software vulnerabilities had previously been used by US authorities and selected companies to close security gaps. In the AI provider's view, however, a limited potential jailbreak does not justify the withdrawal of a commercial model already used by hundreds of millions of people.

Difference Between Fable 5 and Mythos 5

Anthropic announced that it would challenge the order with two lawsuits: one in a federal court in California, another in a federal appeals court in Washington. The company argues that such action must be based on transparent procedures and technical facts. A company-wide application of this standard would, in Anthropic's assessment, effectively stop all new model releases by every frontier provider. Anthropic has so far received only incomplete information from the government.

Legal Pushback from Anthropic

In Germany, the Greens' digital policy politician Süleyman Zorba, currently the Greens' spokesperson on digital policy, warns of the consequences for European digital sovereignty. "Ein einziger Erlass aus Washington, und ein ganzer Kontinent verliert den Zugang zu zentraler Technologie. Keine Mitsprache, kein Einspruch, keine europäische Instanz, die etwas dagegen tun könnte," Zorba said. He added: "Wer seine digitale Infrastruktur vollständig in fremde Hände legt, macht sich abhängig und erpressbar." Zorba also warned: "Was diesmal ein KI-Modell betrifft, könne morgen die gesamte digitale Grundversorgung treffen."

European Reactions and Questions of Sovereignty

Anthropic CEO Dario Amodei had also spoken out just a few days before the order in favor of allowing the government to block potentially dangerous AI software. Now the measure is hitting his own company. Anthropic apologized to its customers for the disruption and said it was working with great urgency to restore access. It is unclear how long the block will last. The company assumes, however, that the authorities have gained knowledge of a method with which the protective mechanisms of Fable 5 could be partially circumvented. For this reason, the company is relying with Fable 5 on a multi-layered security strategy intended to limit potential bypasses and at the same time make them easier to detect.

The deactivation is seen as a litmus test for the AI industry: if a single decree from Washington is enough to take a widely used commercial model offline, this could set a precedent for future disputes between regulatory authorities and developers of powerful AI systems. Observers point out that a significant share of the European economy, public administration, and critical infrastructure relies on cloud services from a small number of US corporations. From this perspective, the block of Fable 5 and Mythos 5 illustrates how vulnerable European structures are in the face of unilateral US export policy. Anthropic itself announced that it would push ahead forcefully with its lawsuits and at the same time seek dialogue with the authorities about possible improvements to the model. What exactly a compromise might look like is currently open.